Fast unpacking and partially packing

[homm]

Benchmarks:

                     Master           This branch
Unpack from L       4697.80 Mpx/s    4745.82 Mpx/s  # No changes
Unpack from LA       615.37 Mpx/s     782.75 Mpx/s
Unpack from RGB      571.12 Mpx/s    1419.82 Mpx/s
Unpack from RGBA    1373.62 Mpx/s    1390.41 Mpx/s  # No changes
Pack to L           2406.48 Mpx/s    2494.54 Mpx/s  # No changes
Pack to LA           727.90 Mpx/s     730.90 Mpx/s  # No changes
Pack to RGB          524.85 Mpx/s     777.51 Mpx/s
Pack to RGBA         669.26 Mpx/s     660.43 Mpx/s  # No changes

[wiredfool]

FWIW, this does pass all the tests on PPC/big endian.

[homm]

Should be rebased after #2694 merge

[homm]

Merged tests from #2694 there.

[homm]

Coverage reduced because number of lines in Unpack.c is reduced.